Machine Learning Can Predict Level of Improvement in Shoulder Arthroplasty

Area of Science:

• Orthopaedic Surgery
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Accurate prediction of postoperative outcomes is crucial in orthopaedic surgery.
• Machine learning (ML) offers powerful predictive modeling capabilities in healthcare.
• Glenohumeral osteoarthritis (OA) impacts shoulder function and necessitates effective treatment strategies.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in predicting functional improvement after shoulder arthroplasty for glenohumeral OA.
• To assess the predictive accuracy of ML models using preoperative data, including patient-reported outcomes and morphological characteristics.
• To determine if ML can predict American Shoulder and Elbow Surgeons (ASES) score improvements at a minimum of two years post-surgery.

Main Methods:

• Retrospective cohort study of 472 patients with glenohumeral OA undergoing shoulder arthroplasty.
• Preoperative CT scans analyzed for glenoid and rotator cuff morphology.
• Three ML models compared: all baseline variables, excluding morphology, and excluding ASES scores.

Main Results:

• The model incorporating all baseline and morphological variables (Model 1) demonstrated the highest prediction accuracy for ASES score improvement.
• Probability values for predicting improvement classes A, B, and C were 0.94, 0.95, and 0.94, respectively, using Model 1.
• Models excluding morphological or ASES variables showed lower, though still significant, predictive accuracy.

Conclusions:

• Machine learning accurately predicts functional improvement following shoulder arthroplasty for glenohumeral OA.
• Combining patient-reported perceptions (ASES scores) with morphological data enhances prediction accuracy.
• Accurate outcome prediction can aid physicians in managing patient expectations and optimizing satisfaction.